A lightweight personal AI assistant framework
Project description
Flowly is an open-source AI agent that runs on your machine, meets you on every channel you already use, and gets better the longer you use it. One shared memory, one library of skills, your own LLM keys. It remembers across conversations, maintains and improves itself over time, schedules its own work, and connects to anything that speaks MCP — from a $5 VPS, a Mac mini, or the laptop in front of you.
Quick start
# Install — one command sets up uv, Python, and Flowly
curl -fsSL https://useflowlyapp.com/install.sh | bash
# (already manage tools with uv? `uv tool install flowly-ai`)
# First-time setup — pick an LLM provider, add any channels
flowly setup
# Start chatting (bare `flowly` opens the terminal UI)
flowly
flowly setup walks you through a provider (OpenRouter, Anthropic, OpenAI, Gemini, Groq, xAI/Grok, Zhipu, or a local model) and any channels you want. Two minutes, done. Add a Telegram bot? flowly setup → Telegram → paste token. The gateway hot-reloads each channel as you save.
What's inside
| One agent, every channel | Terminal TUI · Telegram · Discord · Slack · Microsoft Teams · WhatsApp · iMessage · Email · voice. A single gateway process speaks to all of them, with one conversation memory shared across every surface. Native Mac / iOS / Android apps and a browser extension come via Flowly Cloud (optional). |
| 40+ first-class tools | Workspace-sandboxed file I/O and an allowlisted shell with per-command approval; web search + page fetch + live X/Twitter; macOS computer-use (Accessibility control, screenshots, browser-tab + clipboard); PDF/DOCX/XLSX/PPTX, image generation, video_analyze; Linear, Trello, Home Assistant, and Google Workspace (Calendar, Contacts, Drive, Tasks). |
| A closed learning loop | Every fact you share becomes a governed memory with a calibrated trust score — 👍/👎 to retune it, and a background consolidation pass merges duplicates and retires stale notes. Structured facts also land in a knowledge graph. Opt in and Flowly mines your recurring procedures to write and refine its own skills, every change snapshotted and reversible. → memory · skills |
| 135 built-in skills | Plus skill bundles (one keystroke enables a whole stack) and drop-in Markdown skills. Compatible with the open agentskills.io standard. |
| Delegates and parallelizes | Spawn isolated sub-agents for parallel workstreams (delegate, spawn), run a cross-channel task board sequentially or in parallel (BOARD.md), and hand heavy coding off to a local Codex session (opt-in). |
| MCP, both directions | Connect any MCP server (flowly mcp install …) and run Flowly itself as an MCP server (flowly mcp serve) for Claude Desktop, Cursor, or Claude Code. |
| Scheduled & unattended | Built-in cron — schedule any natural-language prompt and deliver the result to any channel. Run the gateway as a background service that survives reboots. |
| Yours to extend & contain | Full Python plugins (tools, slash commands, channels, lifecycle hooks → PLUGINS.md), switchable personas, and a sandbox (macOS sandbox-exec, Linux bwrap; opt out with FLOWLY_SANDBOX=0). |
Providers — bring your own key
Adapters for OpenRouter, Anthropic, OpenAI, Google Gemini, Groq, xAI / Grok, Zhipu / GLM, any OpenAI-compatible local model (Ollama, LM Studio, vLLM), and a hosted Flowly Cloud option. Sign in to xAI with your SuperGrok / X Premium+ subscription instead of an API key. Switch any time:
flowly # open the chat
/provider openrouter # pick provider
/model anthropic/claude-sonnet-4-6
Or edit ~/.flowly/config.json. When nothing is pinned, Flowly cascades through whatever you've configured (OpenRouter → Anthropic → OpenAI → xAI → …) so it always has a working model.
CLI vs messaging — quick reference
Two entry points: start the terminal UI with flowly, or run the gateway and talk to Flowly from Telegram, Discord, Slack, WhatsApp, iMessage, or email. Once you're in a conversation, most slash commands work the same in both.
| Action | Terminal (TUI) | Messaging channels |
|---|---|---|
| Start chatting | flowly |
flowly setup → add channel, then message the bot |
| New / reset conversation | /new |
/new |
| Change model / provider | /model, /provider |
/model, /provider |
| Switch persona | /persona [name] |
/persona [name] |
| Browse & run skills | /skills or /<skill> |
/<skill> |
| Inspect / correct memory | /memory, flowly memory list |
/memory |
| Sessions & history | /sessions |
/sessions |
| Interrupt current work | Ctrl+C or send a message |
/stop or send a message |
Self-host or cloud
Flowly's agent core is Apache 2.0. Self-host on your laptop, a VPS, or a Mac mini — your keys, your data, your machine. Everything in this repo works with no Flowly account. → self-hosting guide · open source vs. Desktop/Cloud
Optional Flowly Cloud adds the native Mac/iOS/Android apps, cross-device sync, hosted LLM access, and a managed relay that keeps your bot reachable when your laptop sleeps. Sign-in is opt-in (flowly login) and only affects the cloud features; the agent is unchanged.
Common commands
flowly # terminal chat (TUI)
flowly setup # interactive wizard (LLM, channels, integrations)
flowly agent -m "..." # one-shot prompt
flowly service install --start # run the gateway as a background service
flowly restart # smart restart (launchd / systemd / Task Scheduler)
flowly doctor # diagnose config + runtime health
flowly memory list # inspect / correct long-term memory
flowly skill mine --dry-run # preview self-improved skills (opt-in subsystem)
flowly mcp install <name> # add an MCP server from the catalog
flowly mcp serve # expose Flowly as an MCP server
flowly login / logout # Flowly Cloud account (optional)
Other command groups: flowly channels · cron · plugins · skills · bundles · persona · codex · xai · sessions · approvals · pairing. Full reference: flowly --help.
Architecture
The gateway runs as a local daemon, and the whole process runs inside the OS sandbox (sandbox-exec on macOS, bwrap on Linux). Channels route through an in-process message bus; the TUI, desktop, and mobile apps connect as clients over one WebSocket protocol (ws://127.0.0.1:18790) — desktop and mobile can also reach it remotely via the Flowly Cloud relay. Config lives at ~/.flowly/config.json, shared by every client. → internal RPC protocol
Installation paths
| Method | Command | When |
|---|---|---|
| Install script | curl -fsSL https://useflowlyapp.com/install.sh | bash |
Recommended — sets up uv, Python, Flowly, PATH |
uv tool |
uv tool install flowly-ai |
If you already manage tools with uv |
| Source | git clone … && pip install -e ".[dev]" |
Contributors |
After install, run flowly setup. To run without a terminal session open:
flowly service install --start # launchd (macOS) / systemd (Linux) / Task Scheduler (Windows)
It survives reboots and terminal close.
Documentation
- MCP.md — connect external MCP servers + run Flowly as an MCP server
- PLUGINS.md — extend Flowly with custom tools, hooks, commands, skills
- BOARD.md — the cross-channel task board (sequential + parallel agents)
- SECURITY.md — trust model + vulnerability reporting
- SELF_HOSTING.md — run Flowly on your own machine or a VPS
- DESKTOP_VS_OSS.md — what's open source vs. Desktop & Cloud
- CONTRIBUTING.md — dev setup, project layout, PR process
- Engineering deep-dives: memory governance · skill self-improvement · internal gateway RPC
Community
- 🐛 Issues — bugs and feature requests
- 🌐 useflowlyapp.com — apps, hosted cloud, and docs
- 📚 agentskills.io — the open skills standard Flowly speaks
Status
Flowly was open-source in v1.x, closed during the v2 rewrite, and is back open under Apache 2.0. The agent, 40+ tools, 135 skills, and every channel adapter in this repo are the same code that ships inside the Flowly Desktop app — there is no separate "lite" build.
Active development. Issues and PRs welcome.
License
Apache 2.0. Fork it, ship it, embed it. Self-hosted use with your own LLM keys is unrestricted. Commercial use of Flowly Cloud credentials (account / relay / gateway tokens) is governed by our Acceptable Use Policy — broadly: use them via official Flowly clients, not custom harnesses that proxy our subscription.
Acknowledgments
Inspired by Claude Code (Anthropic) and the broader open-source agent community. Built by Nocetic Limited.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file flowly_ai-2.2.5.tar.gz.
File metadata
- Download URL: flowly_ai-2.2.5.tar.gz
- Upload date:
- Size: 2.0 MB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.12.13
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
aa76bf99263e99c894abb83be2328d46cdb7ce9a5d1d7b2c44427cfd96c699c7
|
|
| MD5 |
53aa8345bacff8afe3f58dda52d31c01
|
|
| BLAKE2b-256 |
18230692e300e30863566a9bb562574a3bc2584fbaf5e5aa32529cc04a3569f6
|
File details
Details for the file flowly_ai-2.2.5-py3-none-any.whl.
File metadata
- Download URL: flowly_ai-2.2.5-py3-none-any.whl
- Upload date:
- Size: 2.4 MB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.12.13
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
170642b6546fcd631a7a3ebb0d9e392f5a653498a5872c085ad3ea94d907b0be
|
|
| MD5 |
7129e1b5b0c5252439210482252f9ab1
|
|
| BLAKE2b-256 |
e6b5d5611fdfac2988cf63cfb2487c2cf1b6e094ddd7f5b18f2e002c2a4fcfe9
|